Motion estimation with wavelet transform and the application to motion compensated interpolation

In order to estimate highly reliable motion for image sequences, a novel segment-based motion estimation is described. Segmentation is performed based on zero-crossings of the wavelet transform. Moving objects are extracted from the area of motion using the segmentation result. Multiresolution image decomposition is performed with the biorthogonal wavelet transform, and motion is then hierarchically estimated using the segmentation result. Moving objects and covered/uncovered areas are identified with the segmentation result and the estimated motion information. Simulations were carried out on real image sequence table tennis (352*240 pels).<>